What a Interview Coaching & Mock Interview Platform actually does
Simulates live job interviews with role-aware question generation, real-time transcription feedback, and post-session improvement coaching — all under your agency's brand.
An AI interview coaching platform combines three real-time layers: a foundation LLM generates behavioral, technical, and case questions grounded in the user's uploaded resume and target job description; a streaming speech-to-text engine transcribes answers word-by-word while detecting filler words, pace, and clarity; and a scoring module evaluates answers against STAR or CARL frameworks and produces a written post-session summary with three concrete improvement actions. The audio loop — user speaks, AI transcribes, AI responds — runs in under 90ms with Cartesia Sonic 3.5 as the voice agent, which is borderline acceptable for interview roleplay but will feel slightly mechanical compared to a human.
The market timing is strong: career coaches, MBA programs, bootcamps, and outplacement firms are all searching for a brandable mock interview tool they can resell. Final Round AI serves the B2C segment at $148/mo but has zero white-label infrastructure; Yoodli and Interview Warmup by Google are free or cheap but not rebrandable. That gap — a coach with 50–500 paying candidates who wants to deliver AI prep under their own brand — is exactly what a Lovable MVP or a custom RapidDev build closes. The compliance boundary is the critical differentiator this page must articulate: coaching a candidate about their own preparation sits entirely outside NYC Local Law 144 AEDT scope and EU AI Act Annex III high-risk classification — the moment a customer redeploys the same product as an employer screening tool, they cross both laws simultaneously.
AI capabilities involved
Role-aware interview question generation (behavioral, technical, case)
Real-time speech transcription with filler-word and pace detection
Live AI voice interviewer with sub-90ms latency
STAR/CARL framework answer scoring (advisory only)
Post-session written summary with improvement actions
Who uses this
- Career coaches with 50–500 active clients reselling personalized interview prep under their own brand
- Coding bootcamps bundling mock technical interviews with job placement guarantees
- MBA programs offering structured interview prep as a premium service to students
- Outplacement firms delivering layoff-to-offer coaching for corporate clients at scale
- University career-services offices replacing static practice question banks with live AI simulation
SaaS alternatives on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Final Round AI
Individual job seekers who want AI interview prep without going through a coach
Limited free trial
$148/mo Pro
Pros
- +Real-time AI feedback during live interviews is genuinely impressive and well-executed
- +Strong coverage of behavioral, technical, and case interview formats
- +Browser-based — no install required for candidates
- +Active product with consistent updates
Cons
- −B2C only — no white-label or reseller program whatsoever
- −You cannot brand the experience as your own coaching product
- −No API for coaches to integrate into their existing platforms
- −Per-coach pricing at $148/mo does not scale if you want to serve many candidates under one account
Yoodli
Individual candidates who want speech-pattern feedback without structured interview simulation
Free tier available
$25/mo Pro
Pros
- +Strong real-time speech analytics (pace, filler words, eye contact) at a low price point
- +Clean candidate UX with video review playback
- +Free tier is genuinely functional for individual practice
- +Integration with Zoom for real interview analysis
Cons
- −No reseller or white-label program — you cannot put your brand on it
- −Limited question-bank customization for coaches who want role-specific question sets
- −Analytics are individual-focused — no coach dashboard for managing multiple candidates
- −Free tier is not a business model for an agency reselling at scale
Big Interview
Universities and bootcamps that want a structured, curriculum-tied interview prep product they can license and embed — if branding matters less than content depth
$79/mo individual
Quote-based for institutions
Pros
- +The most curriculum-structured of the category — courses + practice combined
- +Institution licensing exists for universities and bootcamps (quote-based)
- +Strong behavioral question library built by interview coaches
- +Video answer recording and review included
Cons
- −No white-label tier — it is sold as Big Interview, not as your brand
- −Institution pricing requires sales process; not self-serve
- −AI feedback layer is less sophisticated than Final Round AI or Yoodli
- −Not designed for agencies to resell; it is sold direct to institutions
The AI stack
The production pipeline splits into two tiers: a cheap text-only path (GPT-5.4 mini + Supabase) that costs under $0.05/session, and a real-time voice path (Deepgram + Cartesia + LLM) that costs $2.56/hr — the gap between those two cost points is the primary architectural decision.
Question generation
Generates role-appropriate behavioral, technical, and case questions grounded in the candidate's resume and target job description
GPT-5.4 mini
$0.75/$4.50 per M tokensDefault production tier for behavioral and most technical questions
Claude Haiku 4.5
$1.00/$5.00 per M tokensShops already on Anthropic who want a single-vendor AI stack
Gemini 3.1 Flash-Lite
$0.25/$1.50 per M tokensFree-tier MVP or cost optimization at very high volume (10K+ sessions/month)
Our pick: GPT-5.4 mini as the default. Switch to Gemini 3.1 Flash-Lite only above 10,000 sessions per month where cost pressure is real. Do not use flagship models for question generation — the quality ceiling here is content quality, not reasoning depth.
Speech transcription
Converts candidate's spoken answers to text in real-time for analysis and AI response generation
Deepgram Nova-3
$0.0077/min streaming (~$0.46/hr)Any production voice feature where transcription accuracy directly affects coaching quality
Our pick: Deepgram Nova-3 is the only viable choice for real-time coaching feedback. If you are building text-only (skip voice), eliminate this layer entirely and save $0.46/hr.
AI voice agent (optional)
Delivers AI interviewer questions and follow-ups as natural speech with sub-100ms response latency
Cartesia Sonic 3.5
~$35/M characters (~$2.10/hr at interview pacing)Premium interview simulation tier where realistic conversation is the core value proposition
OpenAI gpt-4o-mini-tts
~$0.015/minAsynchronous audio playback of pre-generated question prompts, not live roleplay
Our pick: Cartesia Sonic 3.5 for live voice roleplay tier. OpenAI gpt-4o-mini-tts for async question delivery. Make voice a paid upgrade — don't include it in a free or base tier.
Audio storage
Stores session recordings for playback, self-review, and coach feedback
Cloudflare R2
$0.015/GB/mo storage + $0 egressAny product where candidates replay their recordings multiple times per session
Our pick: Cloudflare R2 is the right default. At 500 users each recording a 30-minute session, you're storing ~15GB/mo — that's $0.22/mo, negligible. Egress would be $0 vs $1.35/GB on S3.
Reference architecture
The pipeline is two parallel flows: a synchronous text path for question generation and scoring, and an asynchronous voice path for real-time roleplay. The hardest engineering challenge is the WebSocket state machine that keeps Deepgram streaming, the LLM generating, and Cartesia speaking in sequence without buffering gaps that break interview immersion.
Candidate uploads resume + selects target role and interview type
Next.js frontend + Supabase storageResume PDF stored in R2; metadata (target role, company, interview type, difficulty) saved to Supabase sessions table. This grounding context is injected into every subsequent LLM prompt.
System generates opening question set (5–8 questions pre-queued)
Supabase Edge Function + GPT-5.4 miniEdge Function calls GPT-5.4 mini with resume text + job description + interview type as context. Returns a JSON array of questions with STAR/CARL tags and expected themes. Pre-queuing avoids cold-start latency at session start.
Candidate begins speaking — Deepgram WebSocket opens
Client-side Deepgram SDK + WebSocket relayBrowser captures microphone stream; client-side Deepgram SDK sends audio chunks to Nova-3 streaming endpoint. Interim transcripts display in real-time. Final transcript fires on end-of-utterance silence detection.
Final transcript triggers AI turn selection
Supabase Edge Function + GPT-5.4 miniEdge Function receives transcript, determines whether to ask a follow-up or advance to next question based on answer completeness score (STAR completeness is measurable via regex + LLM). Returns next AI utterance as text.
AI response synthesized and streamed to candidate
Cartesia Sonic 3.5 (paid tier) / gpt-4o-mini-tts (base tier)Text sent to Cartesia or OpenAI TTS; audio streamed back to browser via ArrayBuffer chunks. Playback begins before synthesis completes (streaming TTS). Total latency from transcript finalize to first audio byte: ~350–500ms.
Session ends — scoring and summary generated
Claude Sonnet 4.6 + Supabase Edge FunctionFull transcript + question list sent to Sonnet 4.6 for post-session analysis. Outputs: STAR score per answer (0–10), top filler words detected, pace analysis, three improvement actions. Stored to Supabase; emailed to candidate.
Coach dashboard surfaces session data
Next.js dashboard + Supabase RLSRow-level security ensures coaches only see their own candidates. Dashboard shows session count, score trends, common filler words, completion rate. No employer-side ranking — this is advisory analytics for the coach.
Estimated cost per request
~$0.02 per question set (GPT-5.4 mini, ~1K tokens out); ~$2.56/hr for real-time voice session (Deepgram $0.46 + Cartesia ~$2.10); ~$0.08 per post-session summary (Sonnet 4.6, ~3K tokens)
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Baseline assumes 200 active candidates each running 2 sessions per month. Toggle between text-only and voice tiers — the cost gap is 50× and determines your pricing floor.
Estimated monthly cost
$52.53
≈ $630 per year
Calculator notes
- Text-only total (200 candidates × 2 sessions): ~$50 infra + $4 LLM = ~$54/mo — very manageable
- Voice total (200 candidates × 2 × 30 min sessions): ~$50 infra + $4 LLM + $420 voice = ~$474/mo — pricing sessions at $15+ is necessary
- Cartesia cost assumes 15 min of AI speech per 30 min session (candidate speaks the other 15 min)
- Deepgram cost does not include diarization add-on; add ~20% if you enable speaker separation
Build it yourself with vibe-coding tools
By Sunday night you can have a working text-based mock interviewer: paste in your resume, select a role, and the AI generates a 5-question behavioral interview, scores your typed answers, and emails you an improvement report. Voice can come in week 2.
Time to MVP
12–16 hours (1 weekend) for text-only; 2–3 weekends for voice
Total cost to MVP
$25 Lovable Pro + ~$40 OpenAI credits + $20 Deepgram trial credits
You'll need
Starter prompt
Build a white-label AI mock interview platform with these features: 1. CANDIDATE ONBOARDING: PDF resume upload (store in Supabase Storage), role selection dropdown (Software Engineer, Product Manager, Data Analyst, Marketing, Finance — customizable), interview type selector (Behavioral, Technical, Case, Mixed), difficulty level (Entry, Mid, Senior). 2. INTERVIEW SESSION: A chat interface that shows AI interviewer questions one at a time. Candidate types their answer. After each answer, the AI gives a brief acknowledgment and asks a follow-up or moves to the next question. Show question number and total (e.g., 'Question 3 of 6'). Store each question + answer pair to Supabase. 3. AI INTEGRATION: Create a Supabase Edge Function at /functions/v1/generate-questions that calls OpenAI GPT-5.4 mini. Prompt: 'You are a senior interviewer at a {role} company. The candidate is applying for {role} at {level} level. Generate 6 interview questions based on this resume: {resume_text}. Return JSON array: [{question, type (behavioral/technical/case), expected_themes: [string]}]'. Create another edge function /functions/v1/score-answer that scores each answer on STAR completeness (0-10) and returns feedback. 4. POST-SESSION REPORT: After all questions are answered, call a /functions/v1/session-summary edge function using Claude Sonnet 4.6 (Anthropic API). Generate: overall score, top 3 strengths, top 3 improvement areas, one concrete action for each improvement area. Display as a clean report and send via Supabase email. 5. COACH DASHBOARD: Protected route /dashboard with Supabase Auth. Shows table of all candidates: name, last session date, average score, session count. Click into candidate to see their session history and reports. Use Recharts for a score-over-time line chart. 6. DATABASE: Supabase tables — candidates (id, email, name, resume_url, target_role, created_at), sessions (id, candidate_id, interview_type, difficulty, started_at, completed_at, overall_score), questions_answers (id, session_id, question, answer, score, feedback), coach_candidates (coach_id, candidate_id). Stack: Vite + React + TypeScript + Tailwind + shadcn/ui + Supabase (Auth + DB + Storage + Edge Functions) + RLS policies on all tables.
Paste this into Lovable
Follow-up prompts (run in order)
- 1
Add Deepgram Nova-3 real-time transcription: open a WebSocket to wss://api.deepgram.com/v1/listen on session start, stream microphone chunks, display interim transcripts in the input field, fire the score-answer edge function on final transcript. Add a microphone button next to the text input — clicking it toggles recording mode. Show a red pulsing dot while recording.
- 2
Wire Cartesia Sonic 3.5 for AI voice: after each question is generated, send the question text to a new edge function /functions/v1/tts-cartesia that calls the Cartesia streaming TTS API. Return audio as an ArrayBuffer, play it in the browser with the Web Audio API. Add a 'Voice Mode' toggle in session settings that enables/disables TTS.
- 3
Add Stripe billing: create two products — Text Coaching ($29/mo, unlimited text sessions) and Voice Coaching ($79/mo, 10 hours voice/mo). Add a /billing route using Stripe Checkout. Store subscription tier in Supabase profiles table. Gate voice features behind the voice-tier check in the session component.
- 4
Build a question bank editor: /dashboard/question-bank route where coaches can add custom questions with tags (role, type, difficulty). These custom questions are injected into the generate-questions prompt for their candidates. Store in Supabase custom_questions table with coach_id.
- 5
Add Illinois BIPA compliance flow: before a candidate starts their first voice session, show a modal with the BIPA consent text (include: purpose of recording, retention period of 90 days, right to request deletion). Require checkbox + email confirmation. Log consent to Supabase bipa_consents table with timestamp and consent text hash. Block voice sessions for candidates without a consent record.
Expected output
By Sunday night: a working text-based mock interviewer where candidates upload their resume, select a role, answer 6 questions, and receive a scored improvement report. A coach can log in and see all candidate sessions. Voice integration adds one more weekend.
Known gotchas
- !Deepgram WebSocket in Lovable requires a custom edge function relay — you cannot open a direct browser-to-Deepgram WebSocket from Lovable's Edge Functions without CORS configuration, which Lovable cannot handle automatically
- !Cartesia Sonic 3.5 returns audio as a stream — Lovable's AI often generates code that tries to load the full audio before playing, which adds 2–5 seconds of perceived latency and defeats the purpose of a fast TTS
- !BIPA (Illinois Biometric Information Privacy Act) applies if you store voice recordings — the written consent + 90-day retention schedule must be implemented before any Illinois candidate uses voice features
- !EU AI Act Art. 50 requires you to disclose the AI interviewer is an AI by August 2, 2026 — add a persistent 'You are practicing with an AI interviewer' banner to the session screen
- !ToS must explicitly restrict employer-side use: if a customer repurposes your coaching product to screen job applicants for their company, they instantly cross into NYC Local Law 144 AEDT territory — add a clause prohibiting this use case
- !Supabase free tier has a 500MB database limit — at 500 candidates × 6 answers × ~500 chars average, you hit 1.5MB quickly; upgrade to Pro ($25/mo) before inviting real users
Compliance & risk reality check
Interview coaching is the rare HR page where the compliance load is genuinely light — coaching a candidate about their own preparation sits outside the major HR AI laws — but three specific triggers can flip that status fast.
Illinois AI Video Interview Act — employer-side screening trigger
The Illinois AI Video Interview Act (effective January 1, 2020) requires employer consent + candidate notice + race/ethnicity reporting when an employer uses AI to analyze video interview responses for hiring decisions. Your coaching product dodges this because the user is self-coaching — but the moment a customer redeploys the platform to screen candidates on behalf of their company, all three obligations activate. This is a product liability risk, not a personal risk.
Mitigation: Add explicit ToS language prohibiting employer-side screening use. In your onboarding flow, ask the account type (coach/bootcamp/individual) and block employer-account types from accessing session scoring. Consult counsel before building any 'share my results with recruiters' feature.
BIPA — Illinois Biometric Information Privacy Act
If you store voice recordings of users, BIPA applies to Illinois residents. BIPA requires: written informed consent before collecting biometric data (which includes voiceprints), a publicly available retention policy, and destruction of biometric data within 3 years or when the purpose is fulfilled. Violations carry $1,000–$5,000 per violation — and BIPA has a private right of action that has produced class-action suits.
Mitigation: Before any Illinois user starts a voice session, display a BIPA consent modal with explicit retention terms (90 days max for coaching recordings). Store consent records in your database with timestamp and consent text hash. Honor deletion requests within 30 days. Log and audit compliance. If you operate voice recording at scale, consult a privacy attorney.
EU AI Act Art. 50 — chatbot disclosure
From August 2, 2026, EU AI Act Art. 50 requires that any AI system designed to interact with people disclose that the user is interacting with AI. The requirement applies to both the provider (you) and deployers (your coach customers). A mock interviewer that speaks or types as a 'Senior Interviewer at Google' without disclosure would violate this obligation for any EU user.
Mitigation: Add a persistent on-screen disclosure during every session: 'You are practicing with an AI interviewer, not a real person.' Ensure this cannot be disabled by coaches. The disclosure needs to be in the same language as the session. Legacy systems have until December 2, 2026 under the May 7, 2026 EU AI Act Omnibus deal.
AB 2013 — California training-data summary
California AB 2013 (effective January 1, 2026) requires generative-AI developers serving Californians to publish a summary of training data used. This applies to you only if you fine-tune models on user session recordings — not if you use OpenAI or Anthropic APIs without fine-tuning.
Mitigation: Do not fine-tune foundation models on user recordings without publishing a training-data summary. Use API-tier provider calls (which exclude your data from training by default) for all AI features. If you ever fine-tune, publish the required summary at your product URL.
Build vs buy: the real math
8–12 weeks
Custom build time
$13,000–$25,000
One-time investment
6–10 months
Breakeven vs buying
At 500 active candidates paying $49/mo each, the platform generates $24,500/mo. At that scale, the cheapest white-label alternative (Final Round AI at $148/mo per coach account, not per candidate) doesn't support multi-tenant coaching economics at all — you'd need a completely different product category. A custom RapidDev build at $13K–$25K recoups in 1–2 months at 500 paying candidates, and you keep 100% of the margin as Deepgram and Cartesia pricing continues to fall. The real breakeven question is voice: if you build voice into V1, the $13K–$25K estimate climbs to $20K–$35K (real-time WebSocket architecture is the hardest engineering component in this stack). If you start text-only and add voice in V2, you stay in the standard band and ship 4–6 weeks faster. Model prices are dropping: Cartesia's per-character rate has fallen ~40% in 12 months, which means the $2.56/hr voice cost will likely be $1.50/hr by mid-2027 — improving your per-session margins without any product changes.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship.
Discovery call (free)
30 minWe map your exact Interview Coaching & Mock Interview Platform use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
8–12 weeksOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
8–12 weeks
Investment
$13,000–$25,000
vs SaaS
ROI in 6–10 months
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does it cost to build a white-label AI interview coaching platform?
A text-only MVP on Lovable costs $25 (Lovable Pro) + ~$40 in API credits for a weekend build. A production multi-tenant platform with voice costs $13,000–$25,000 with RapidDev (8–12 weeks). Adding real-time voice to the custom build — Deepgram WebSocket + Cartesia TTS — pushes the estimate toward $20K–$35K. The biggest cost variable is voice: text-only sessions run about $0.10 each; voice sessions run $2.56/hr in API spend.
How long does it take to ship a mock interview platform?
A text-only Lovable MVP can be running by Sunday night — 12–16 hours of work. Adding Deepgram real-time transcription takes another weekend. A full production platform with voice, multi-tenant coach dashboard, Stripe billing, and BIPA compliance flow is an 8–12 week RapidDev engagement. The voice WebSocket state machine is the longest engineering task (2–3 weeks alone).
Does interview coaching AI fall under NYC Local Law 144 or the EU AI Act?
No — if the product coaches candidates about their own preparation. NYC Local Law 144 AEDT applies when an employer uses AI to make or substantially assist in hiring, promotion, or termination decisions. A coaching tool where the candidate is the user making decisions about their own practice sits outside this scope. The same exemption applies under EU AI Act Annex III, which covers employment decisions made by employers, not self-coaching by candidates. The critical watch-out: if a customer repurposes your platform to screen candidates on behalf of their company, all those laws activate instantly. Your ToS must explicitly prohibit employer-side screening use.
Is real-time voice really necessary, or is text-only good enough?
Text-only is genuinely good enough for most coaching use cases. The vast majority of interview prep value comes from question quality, answer scoring, and improvement feedback — none of which requires voice. Real-time voice adds realism and is a meaningful differentiator for executive coaching or high-stakes interview prep, but it costs 50× more than text-only (~$2.56/hr vs ~$0.05/session). The practical recommendation: launch text-only, validate that candidates pay for it, then add voice as a premium tier.
Can RapidDev build this for my coaching business?
Yes. RapidDev has built 600+ applications including AI coaching and career-tech products. A typical engagement for a white-label interview coaching platform runs $13,000–$25,000 over 8–12 weeks, covering multi-tenant coach dashboard, Supabase backend, question generation, post-session scoring, Stripe billing, and BIPA compliance flow. Voice integration is scoped separately at $5K–$10K additional. Book a free 30-minute consultation at rapidevelopers.com to scope your specific requirements.
What is Illinois BIPA and does it apply to my interview coaching app?
Illinois Biometric Information Privacy Act (BIPA) applies when you collect, store, or use biometric identifiers — which includes voiceprints derived from audio recordings. If any of your users are Illinois residents and you store their voice recordings, BIPA requires written informed consent before collection, a published retention policy, destruction of data within 3 years, and prohibition on selling biometric data. BIPA has a private right of action with $1,000–$5,000 per violation that has produced major class-action suits. If you use voice features, implement a consent modal before the first voice session and store consent records with timestamps.
What is the cost breakdown for voice vs text-only tiers?
Text-only: ~$0.02/session for question generation (GPT-5.4 mini, ~1K tokens) + ~$0.08/session for post-session summary (Claude Sonnet 4.6, ~3K tokens) = ~$0.10/session total. Voice: add ~$0.46/hr Deepgram Nova-3 transcription + ~$2.10/hr Cartesia Sonic 3.5 TTS = ~$2.56/hr in voice API alone. A 30-minute voice session costs about $1.28 in voice API. At a $30/mo subscription, you can support about 23 text-only sessions or about 23 minutes of voice per user before margin disappears.
Want the production version?
- Delivered in 8–12 weeks
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.